Learning investment · first-year return

ROI Calculator

Understand how learning creates financial value, build a starting estimate of first-year ROI, and identify what to measure as your program progresses. Start with an example, then replace its figures with your own. The calculation follows the Phillips ROI method (ROI Institute, 2019).

Scope: financial ROI in the first year

ROI compares the estimated dollar value of benefits with the costs you enter, including learners' time. It does not prove that learning caused a change.

OutcomeIncluded?Why & how
Time saved and put to useful workYesValued at employment cost, or extra profit per useful hour. Time value is not automatically a cash saving.
Fewer people leavingYesEstimated leavers avoided × replacement cost.
Cash savings or additional profitYesEnter under Other measurable value, per person applying it, per year. Count each benefit once.
Capability, performance ratings and promotionsNoUseful evidence of progress, but not dollar benefits on their own. Consider alongside the financial estimate.
Benefits after year oneNoThis provides a first-year baseline; later benefits need a separate estimate.
Example scenarios:
What this scenario assumes

–

–

Starting figures and sources

    These are constructed examples using figures from different sources, not programs evaluated by those studies. The sources explain the starting inputs; they do not establish the calculated ROI.

    ROI, year one–(Benefit − cost) ÷ cost
    Return per $1–Estimated benefit ÷ cost
    Break-even use rate–Share of participants who must apply the learning for year-one benefits to equal total cost–

    How the value builds

    Change a figure to see its effect. Boxes grow or shrink with the size of their value compared with the starting example; size changes are limited to keep labels readable. Click a box to bring its explanation into view.

    –
    –
    –

    1. Program and people

    What you spend, and whose time it takes
    Number attending your program. This determines whose learning time is costed and how many people may apply it.
    A$
    Design, delivery, facilitation, champions and evaluation, excluding learners' time and licences. The examples use ATD's average cost per learning hour used of US$165 in 2024 (ATD, 2025, p. 17), converted at A$1.418 per US$ (RBA, 2026) and multiplied by people × learning hours. This is a US benchmark; replace it with your quoted costs. The budget stays fixed when you change people or learning hours.
    A$/yr
    Year-one software or platform costs the benefits depend on. The ROI method counts all program costs, including software (ROI Institute, 2024, p. 21), so a tool's benefits are counted together with its cost. The AI example uses Microsoft 365 Copilot Business at A$31.40 per user per month, paid yearly, excluding GST (Microsoft, 2026): 300 × A$31.40 × 12 = A$113,040. Where a licence serves several programs, include a fair share.
    hrs
    Hours each participant spends learning. Their time is included in the total cost.
    A$
    A$2,083.70 a week × 52: average full-time adult ordinary time earnings, May 2026 (ABS, 2026a). Managers usually earn more; enter your participants' average where you know it.
    %
    12% super guarantee from 1 July 2025 (ATO, 2025). Payroll tax, workers' comp and leave loading are not included
    hrs
    Planning assumption: 38 hrs × 44 working weeks. It subtracts 4 weeks' annual leave, 10 days' personal leave and about 10 public holidays, the entitlements in the National Employment Standards (Fair Work Ombudsman, n.d.). Personal leave is an entitlement and many people take less, so replace this with your own attendance data where available. Weeks in this model = annual hours ÷ 38.
    –

    2. Use on the job

    Time and performance gains depend on using the learning
    %
    Average share applying the learning across year one. Training professionals in 150 organisations reported 62% applying training immediately, 44% at six months and 34% at one year (Saks & Belcourt, 2006). The examples use 46%, the year-one average of those three points assuming a straight-line decline between them. These are professionals' estimates for training generally; measure use in your own program.

    3. Performance gain

    Enter the benefits relevant to your program; 0 adds no benefit
    hrs
    Enter a value to include time savings. Use hours per person applying the learning. The AI example uses 2.0 hrs: 5.2% of a 38-hour week, the average time saved reported by 925 US workers who used generative AI for work in November 2024 (Bick et al., 2026). Self-reported savings provide context; use an estimate for your own users.
    %
    Share of saved hours that become useful work. The AI example uses 61.2%: the share of freed-up time put back into work, reported by 381 Japanese employees whose work time fell with generative AI (PERSOL Research and Consulting, 2026, p. 40). Self-reported and context-specific; replace it with your own evidence. Read the study.
    A$/hr
    Use additional revenue less the extra costs of earning it, for hours you expect to sell. The example rate is illustrative.
    A$/yr
    Enter a value to include this benefit. Annual value per person applying it: cash savings, additional profit or other gains you can value in dollars. Exclude benefits already counted under time or retention. For a total annual benefit, divide by the expected people applying it before entering it here.
    pts
    Enter a value to include retention savings: the expected fall in annual turnover, in percentage points. For example, 10% leaving before and 8% after is a 2-point drop. The leadership example uses 1.1 points: Australia's 7.2% annual job mobility (ABS, 2026b) × 15%. The 15% is the only estimate from a trial of leadership training itself: trained supervisors were 15% less likely to quit (Adhvaryu et al., 2023). Higher figures of 20% to 40% come from trials of different interventions: a CEO asking store managers to cut quits in a retail chain where about 90% of staff left each year, with the effect fading after nine months (Friebel et al., 2022), and workplace-climate training for whole head offices, where leaders were encouraged to take part (Alan et al., 2023). The training trial also found no detectable change in team turnover, and this model applies the 15% to teams as well, so a higher figure would stretch the evidence further. Replace it with your own turnover rate and evidence; enter 0 to leave retention out. Count participants and their teams as distinct people.
    For leadership retention, team size determines how many employees may benefit from each participant applying the learning. The example uses six, Gallup's median of about five to six direct reports per US manager in 2025; the mean of 12.1 reflects a minority of very large teams (Harter, 2026). Use your own average, or 0 to count participant retention only.
    A$
    Default: average full-time adult ordinary time earnings (ABS, 2026a)
    % pay
    Median 21% of annual salary across US case studies, excluding executives and physicians (Boushey & Glynn, 2012). Senior roles cost more to replace, with some estimates up to 213%; Gallup gives half to two times salary without a stated method (McFeely & Wigert, 2019).

    4. Allow for other causes and uncertainty

    Keep the share linked to learning, then allow for uncertainty
    %
    How much is due to this program rather than changes in tools, management or market conditions? 70% keeps 70 cents of each dollar of estimated benefit. The 70% comes from an ROI Institute worked example (2024, p. 16). When a benefit comes from a trial with a control group, that comparison already isolates the program's effect, so the leadership example uses 100%.
    %
    80% keeps 80% of the value remaining after the first adjustment, allowing 20% for uncertainty. It is a judgement, not a statistical confidence level, taken from the same ROI Institute worked example (2024, p. 16). In the leadership example it allows for applying a trial from another country and industry to your workplace. If a benefit already allows for these factors, do not deduct them again.

    What this tells you

      ROI by how many people apply it

      This changes the share applying it while holding all other inputs fixed. The dot marks your estimate. Above zero, estimated benefits cover costs in year one.

      First-year ROI against the share of people applying the learning Line chart. ROI rises in a straight line as more people apply it.

      How the numbers add up

      Where the defaults come from

      Starting figure is a preset value. Your figure is a value you entered. The notes explain which figures come from research, which are worked examples, and which describe the program. A research source does not establish that a figure fits your workplace. Zero means no benefit is included yet; enter your estimate to include it.

      Starting figures for this example (your edits appear above): –

      AssumptionDefaultStatusBasis

      References